What Turnitin AI score is safe for an MD/MS/DNB thesis?

Most Indian medical universities have not yet published an official Turnitin AI-writing threshold, but examiners and PG sections in 2026 are flagging theses with AI scores above roughly 20–25%. Under 10% is the safe target most guides and Glomerulus aim for. Always confirm with your university's PG section before submission.

Is the Turnitin AI score the same as the plagiarism similarity score?

No. They are two separate reports. The similarity index measures overlap with previously published text; most Indian universities cap this at 10% (some allow up to 15–20%). The AI score estimates how much of the text appears to be written by generative AI such as ChatGPT, and is reported separately. A thesis can pass similarity but still fail the AI check.

Which Indian medical universities have a defined Turnitin similarity limit?

The Tamil Nadu Dr. M.G.R. Medical University, RGUHS Karnataka, MUHS Maharashtra and most NMC-affiliated colleges commonly use a Turnitin similarity limit of 10% (excluding references and quotes). DNB theses under the NBE follow similar conventions. AI-writing limits are newer and are being added to PG handbooks in 2026 — check yours for the latest version.

My thesis has a 40% AI score. Will it be rejected?

It depends on the university and the examiner, but a 40% AI score is well above what most Indian PG sections accept in 2026. The safer route is to bring it under 10% before submission. A doctor-led rewrite preserves your study's substance, references and statistical results while removing AI signatures from the prose.

How do I reduce a high Turnitin AI score?

Rewrite AI-generated passages in your own clinical voice, vary sentence rhythm, replace generic phrases with specific case details from your dataset, and keep all references and numerical results intact. AI paraphrasing tools rarely clear modern detectors. A doctor-led human rewrite is the most reliable route — Glomerulus offers this from ₹500 per chapter.

Does Turnitin also detect ChatGPT in references and methods sections?

Turnitin's AI report scans the body text. Vancouver references, drug names, p-values and statistical output are normally not flagged. The risk areas are introduction, review of literature and discussion — chapters where residents most often use ChatGPT to draft prose.

Turnitin AI Score Limit for an MD/MS/DNB Thesis (2026)

In 2026, Indian medical universities run two separate Turnitin reports on every postgraduate thesis: a similarity check (the classic "plagiarism percentage") and an AI-writing check (the newer ChatGPT detector). Most residents know the 10% similarity rule. Far fewer realise that a Turnitin AI score of 40% — common in theses where ChatGPT drafted the review of literature or discussion — can stall the viva even when similarity is perfectly clean. This guide explains both numbers, what each university actually accepts, and how to bring a high AI score down to a safe level without losing your references or your study's substance.

The two scores: similarity vs AI-writing

The Turnitin report you submit with your thesis contains two distinct metrics, and they measure completely different things.

Similarity index — percentage of your text that overlaps with previously published material (journal articles, theses, websites). This is the traditional "plagiarism" number. Indian universities have capped it for years.
AI-writing score — Turnitin's estimate of how much of your text appears to be generated by a large language model such as ChatGPT, Claude or Gemini. Rolled out across academic Turnitin in 2023, and increasingly used by Indian PG sections since 2024.

A thesis can pass one and fail the other. Heavy paraphrasing of published reviews tends to push similarity up and AI down. Asking ChatGPT to write your discussion from scratch usually does the opposite — similarity is low because the text is novel, but the AI score climbs above 40%.

The single most useful thing to know in 2026: when an examiner says "your Turnitin is high," ask which number. Fixing one does not fix the other, and the rewriting strategy for each is different.

What Indian medical universities accept in 2026

Most Indian medical universities now circulate written limits for similarity. Limits for AI-writing are newer and less uniformly documented — they are appearing in PG handbooks through 2025–26, so always confirm the latest version with your university's PG section before submission. The figures below are the working thresholds used by examiners and Glomerulus' own thesis evaluations across hundreds of residents.

University / body	Similarity limit	Working AI-score limit (2026)
The Tamil Nadu Dr. M.G.R. Medical University	10% (excluding references)	Aim < 10%; flagged > 25%
RGUHS Karnataka	10%	Aim < 10%; flagged > 25%
MUHS Maharashtra	10–15%	Aim < 10%; flagged > 25%
NMC-affiliated colleges (general)	10% (commonly)	No NMC-wide rule yet; safest target < 10%
NBE / DNB	10%	Aim < 10% — NBE is increasingly scrutinising AI content

If your university handbook does not yet mention an AI-writing limit, the operating rule from examiners and external reviewers in 2026 is simple: keep both numbers under 10%. That is the threshold at which a thesis sails through; above 20–25% on either, expect the file to be returned for revision before the viva.

Common reasons MD/MS/DNB theses fail the AI check

Across 800+ theses Glomerulus has reviewed, four patterns produce almost every high AI score:

ChatGPT-drafted discussion. Residents paste their results into ChatGPT and ask for a discussion. The output reads fluently and triggers AI detectors immediately — the rhythm, transition phrases and even-handed tone are characteristic.
AI-rewritten review of literature. A common shortcut is feeding a previous thesis or a Cochrane review into an AI tool to "rewrite in different words." Detectors flag this even when similarity drops.
AI-translated text. Drafting in another language and machine-translating into English creates the smooth, slightly generic prose AI detectors associate with LLM output.
AI-paraphrased methods. The methods section is mostly factual and should match your protocol — passing it through Quillbot or ChatGPT often raises the AI score without reducing similarity.

The discussion and review of literature are where most of the AI score lives. The introduction, results, references and annexures rarely contribute.

How to bring a high AI score under 10%

AI paraphrasing tools rarely clear modern detectors — they swap synonyms while keeping the underlying sentence structure, which is exactly what the detector keys on. The reliable route is a human rewrite that does four things at once:

Rebuild the sentence rhythm. Vary sentence length, break up the long, balanced sentences AI loves, and let the prose run uneven the way human academic writing does.
Insert clinical specificity from your dataset. Replace generic phrases ("our study found a significant association") with specifics your data actually shows ("among 142 patients, the mean HbA1c was 8.6 ± 1.4 and 38% had retinopathy"). AI tools cannot generate these — only you and your masterchart can.
Keep all references, p-values and drug n